search

nerfstudio-project / nerfstudio

A collaboration friendly studio for NeRFs
https://docs.nerf.studio
Apache License 2.0
9.35k stars 1.27k forks source link

"Error: Unknown compute capability. Ensure PyTorch with CUDA support is installed." (using docker) #1731

Open sabinecelina opened 1 year ago

sabinecelina commented 1 year ago

I am building the Docker container using Dockerfile. However, when I run the Docker container, I encounter the following error:

❌ Completion script generation failed: ['ns-train', '--tyro-print-completion', 'bash']

import tinycudann as tcnn
             File "/home/user/.local/lib/python3.10/site-packages/tinycudann/__init__.py", line 9, in
           <module>
               from tinycudann.modules import free_temporary_memory, NetworkWithInputEncoding, Network,
           Encoding
             File "/home/user/.local/lib/python3.10/site-packages/tinycudann/modules.py", line 16, in
           <module>
               raise EnvironmentError("Unknown compute capability. Ensure PyTorch with CUDA support is 
           installed.")

I have confirmed that the NVIDIA CUDA Toolkit is installed in my Windows Subsystem for Linux 2 (WSL2) environment.

Can anyone provide insights on how to resolve this issue and successfully run PyTorch with CUDA support in my Docker container?

TaoF02 commented 1 year ago

hello, I encountered the same error and have sloved by this: https://github.com/nerfstudio-project/nerfstudio/issues/1056#issuecomment-1412152798

sabinecelina commented 1 year ago

I solved this problem by using the --privileged flag for Docker, as my GPUs have other processes running that block the GPU. So the main bug was that even nvidia-smi would not run in the nvidia-base image that nerfstudio was using.